A self-evolving fake news detection method for social media

Through the mechanism of hard attention and multi-head self-attention, combined with deep learning models, effective detection of false messages in new events is achieved, the problem of poor continuous learning ability in the existing technology is solved, and the self-evolution ability of false message detection is improved.

CN115952362BActive Publication Date: 2025-04-04NORTHWESTERN POLYTECHNICAL UNIV
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202310003377.1
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2023-01-03
Publication Date
2025-04-04
Estimated Expiration
2043-01-03

AI Technical Summary

Technical Problem

Existing fake news detection methods are difficult to effectively detect fake news in new events, and have poor continuous learning ability, making it difficult to adapt to changing news events on social media.

Method used

The self-evolving false message detection method is adopted, and the knowledge storage mechanism of hard attention and the knowledge transfer mechanism of multiple self-attention is used to extract post features and store historical event knowledge to realize the classification of false messages in new events.

Benefits of technology

Continuous detection of fake messages on social media is realized, and the model is trained from scratch without storing historical data, which improves the self-evolution and adaptability of fake message detection methods.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN115952362B_ABST
    Figure CN115952362B_ABST
Patent Text Reader

Abstract

The present invention discloses a self-evolving fake news detection method for social media, which collects social media data, uses a deep learning model to extract features to train a classifier, and uses the learned similar historical event knowledge stored in a feature similarity transfer model to help train a fake news classifier for new events, thereby achieving continuous fake news detection. The invention is based on two core mechanisms: one is a knowledge storage mechanism based on hard attention, in which the memory unit is used to store historical event knowledge, and the event mask is used as a unique identifier for each event in the memory unit, and the historical event knowledge can be called at any time; the other is a knowledge transfer mechanism based on multi-head self-attention, which fuses historical event knowledge to improve the fake news detection effect of the classifier on the current event. The present invention can continuously detect fake news on social media without storing historical data to train the model from scratch, thereby improving the self-evolving adaptive ability of the fake news detection method.
Need to check novelty before this filing date? Find Prior Art

Citation Information

Patent Citations

  • False message early detection method based on crowd-sourcing data fusion

    CN110765108A

  • Method for detecting false messages in social media

    CN112100515A